In the last couple of years, neo-banking has started making waves in the fintech industry. Neobanks exist exclusively on the internet and cover all the traditional banking needs such as payments, lending, wallet, insurance etc.
PayU, founded in 2002, offers fintech technology and payment gateway solutions. Sachin Garg, Head of Data Science at PayU, spoke about the company’s services, use of machine learning, and the ‘tech’ powering the financial industry at MLDS 2021.
About PayU
“The fintech industry was among the earliest proponents of data science. Of course, the algorithms were limited, but their use was still seen in the fintech industry from a much earlier time for applications as vital as to whether a customer must be given loan or not,” said Garg.
PayU is fully owned by a $100 billion group named Prosus, which also holds stakes in leading companies such as Swiggy and Byjus. PayU has a presence in over 18 countries and is a leading provider in 36 markets.
PayU India, in particular, has two main sections–PayU Payments and PayU Finance.
Wealth of Data
As per Garg, data is the cornerstone of the fintech industry. With 200,000 merchant partners across companies such as Facebook, Netflix, Flipkart, and Myntra, he claimed, PayU has the largest depth of data in the digital space in India.
“The most exciting features of PayU is having data both on the lending and the spending side. This helps us to chart customer’s end-to-end data lifecycle, helping us to take critical decisions in terms of lending, risk behaviour and assessment, etc,” said Garg.
PayU Finance deals with third party data, bureau data, personal data, financial data, mobile data, and graph data.
PayU’s data science team uses state-of-the-art machine learning algorithms to cull insights from the available database. The prominent models include: propensity model; income/affluence model; risk/fraud model, process automation; collection models; and personalisation.
Speaking of a popular ML use case in place at PayU, Garg said, “There are customers who take a loan from us–it could be a Swiggy loan or a large cash loan. Sometimes they are not able to pay back within the deadline. Most times, people who miss paying on deadline payback in the next 7 days. Our ML model helps us predict how likely a customer will pay back at least 7 days after the deadline. This kind of data and estimation helps us reduce efforts on regular reminders and callbacks. And instead, we can concentrate our efforts on other defaulters. It is a win-win situation for both the customer and the company.”
Another use case is customer service automation. “We have built an inhouse chatbot as the domain is very specific,” said Garg. Here, the ML models are faced with a range of challenges that include multi-stage loan lifecycle, a broad range of queries, and query language.
Customer queries come in English, Hinglish (Hindi and English), and in Hinglish with mistakes. “Our NLP model considers these constraints, understands query correctly and then classifies it into more than 350 intents. Combining the intent classification with predetermined rules, we provide instantaneous answers to these customer queries.”
Garg spoke in length about the use of data science at PayU:
- The data science and the data platform at the company is under the same umbrella, enabling faster model deployment and lesser dependency on core tech teams
- Being part of the larger Prosus ecosystem helps in access to the global data scientist community
- There is a great focus on fair and trustworthy AI
